Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

CPP: Fachverband Chemische Physik und Polymerphysik

CPP 25: Poster Session II

CPP 25.28: Poster

Dienstag, 2. April 2019, 14:00–16:00, Poster B1

Prediction of polymeric nano-structures via machine learning — •Lucia Wesenberg and Ludwig Schneider — Institute for Theoretical Physics, University Göttingen, Germany

The significant length and times scales of the self-assembly of copolymers pose a challenge to particle-based polymer simulation. One possible speed-up strategy consists of using the chemical potential of a non-equilibrium morphology to predict the time evolution. The calculation of the chemical potential by particle-based simulation, however, can take up a vast amount of time, and here we explore the use of machine learning to predict the chemical potential. Machine learning has gained importance due to the introduction of deep neural networks. These enable an efficient implementation of non-linear relations.

Here, we employ this technique to calculate the chemical potential of copolymers in the lamellar phase. Data from different models enable us to tackle the particular interactions separately. First, with data from the Swift-Hohenberg model, we implement the short-range interactions in the neuronal network. Then, we amend the architecture of the network to cover long-range interactions. These are trained using data obtained from the Otha-Kawasaki model. Additionally, a network structure independent of the size of the input data is desirable.